Differential Neighborhood Selection In Memory-Based Group Recommender Systems

As recommender systems have become commonplace to support individual decision making, a need has also been recognized for systems that tailor and provide recommendations to a group of users together rather than individuals alone. Group recommender research to date has focused on evaluating strategies for aggregating profiles of group members to form a consolidated group profile or for aggregating recommendations to individual group members as a consolidated group recommendation list. This paper presents a novel neighborhood selection approach for group recommendation in the context of a neighborhood-based Collaborative Filtering system. We evaluate the performance of this approach with respect to group characteristics such as size and group member similarity. Results show that this approach can result in more accurate predictions for the group, particularly for groups that are more homogenous.